In [ ]:
### Initial dataset visualization with histograms
for c in data.columns:
sns.FacetGrid(data,
hue="Attrition",
height= 5).map(sns.histplot,c).add_legend()
### Imports
import mrmr
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from sklearn.preprocessing import PolynomialFeatures
from sklearn.feature_selection import RFECV
from sklearn.linear_model import LogisticRegression
### Import data
data = pd.read_csv("train.csv")
# Get general info
print(data.info(), "\n\n\n")
print(data.describe(), "\n\n\n")
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1340 entries, 0 to 1339
Data columns (total 35 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 EmployeeID 1340 non-null int64
1 Age 1340 non-null int64
2 Attrition 1340 non-null object
3 BusinessTravel 1340 non-null object
4 DailyRate 1340 non-null int64
5 Department 1340 non-null object
6 DistanceFromHome 1340 non-null int64
7 Education 1340 non-null int64
8 EducationField 1340 non-null object
9 EmployeeCount 1340 non-null int64
10 EnvironmentSatisfaction 1340 non-null int64
11 Gender 1340 non-null object
12 HourlyRate 1340 non-null int64
13 JobInvolvement 1340 non-null int64
14 JobLevel 1340 non-null int64
15 JobRole 1340 non-null object
16 JobSatisfaction 1340 non-null int64
17 MaritalStatus 1340 non-null object
18 MonthlyIncome 1340 non-null int64
19 MonthlyRate 1340 non-null int64
20 NumCompaniesWorked 1340 non-null int64
21 Over18 1340 non-null object
22 OverTime 1340 non-null object
23 PercentSalaryHike 1340 non-null int64
24 PerformanceRating 1340 non-null int64
25 RelationshipSatisfaction 1340 non-null int64
26 StandardHours 1340 non-null int64
27 Shift 1340 non-null int64
28 TotalWorkingYears 1340 non-null int64
29 TrainingTimesLastYear 1340 non-null int64
30 WorkLifeBalance 1340 non-null int64
31 YearsAtCompany 1340 non-null int64
32 YearsInCurrentRole 1340 non-null int64
33 YearsSinceLastPromotion 1340 non-null int64
34 YearsWithCurrManager 1340 non-null int64
dtypes: int64(26), object(9)
memory usage: 366.5+ KB
None
EmployeeID Age DailyRate DistanceFromHome Education \
count 1.340000e+03 1340.000000 1340.000000 1340.000000 1340.000000
mean 1.460265e+06 36.580597 799.197761 9.193284 2.924627
std 2.494821e+05 9.013072 399.333256 8.141621 1.036088
min 1.025177e+06 18.000000 102.000000 1.000000 1.000000
25% 1.237599e+06 30.000000 465.000000 2.000000 2.000000
50% 1.469862e+06 35.000000 796.000000 7.000000 3.000000
75% 1.670131e+06 42.000000 1153.000000 14.000000 4.000000
max 1.886378e+06 60.000000 1499.000000 29.000000 5.000000
EmployeeCount EnvironmentSatisfaction HourlyRate JobInvolvement \
count 1340.0 1340.000000 1340.000000 1340.000000
mean 1.0 2.709701 65.559701 2.717910
std 0.0 1.099961 20.335025 0.717523
min 1.0 1.000000 30.000000 1.000000
25% 1.0 2.000000 48.000000 2.000000
50% 1.0 3.000000 65.000000 3.000000
75% 1.0 4.000000 83.000000 3.000000
max 1.0 4.000000 100.000000 4.000000
JobLevel ... RelationshipSatisfaction StandardHours Shift \
count 1340.000000 ... 1340.000000 1340.0 1340.000000
mean 2.051493 ... 2.700000 80.0 0.808209
std 1.104491 ... 1.079858 0.0 0.856251
min 1.000000 ... 1.000000 80.0 0.000000
25% 1.000000 ... 2.000000 80.0 0.000000
50% 2.000000 ... 3.000000 80.0 1.000000
75% 3.000000 ... 4.000000 80.0 1.000000
max 5.000000 ... 4.000000 80.0 3.000000
TotalWorkingYears TrainingTimesLastYear WorkLifeBalance \
count 1340.000000 1340.000000 1340.000000
mean 11.222388 2.785821 2.771642
std 7.696043 1.263473 0.700007
min 0.000000 0.000000 1.000000
25% 6.000000 2.000000 2.000000
50% 10.000000 3.000000 3.000000
75% 15.000000 3.000000 3.000000
max 40.000000 6.000000 4.000000
YearsAtCompany YearsInCurrentRole YearsSinceLastPromotion \
count 1340.000000 1340.000000 1340.000000
mean 7.070149 4.272388 2.175373
std 6.039663 3.677798 3.222376
min 0.000000 0.000000 0.000000
25% 3.000000 2.000000 0.000000
50% 5.000000 3.000000 1.000000
75% 10.000000 7.000000 3.000000
max 40.000000 18.000000 15.000000
YearsWithCurrManager
count 1340.000000
mean 4.167164
std 3.581605
min 0.000000
25% 2.000000
50% 3.000000
75% 7.000000
max 17.000000
[8 rows x 26 columns]
### Initial dataset visualization with histograms
for c in data.columns:
sns.FacetGrid(data,
hue="Attrition",
height= 5).map(sns.histplot,c).add_legend()
# Make pairwise scatter plots promising features seen in above analysis
promising_features = ["Age", "DistanceFromHome",
"EnvironmentSatisfaction",
"JobLevel", "JobRole",
"MaritalStatus",
"NumCompaniesWorked", "OverTime",
"Shift", "YearsAtCompany",
"YearsInCurrentRole",
"Attrition"]
promising_data = data[promising_features]
sns.pairplot(promising_data,
hue="Attrition", height= 3)
<seaborn.axisgrid.PairGrid at 0x2abee75b0>
# Import other data
labels = data.pop("Attrition")
submission_data = pd.read_csv("test.csv")
print(data.shape)
print(submission_data.shape)
(1340, 34) (336, 34)
# drop columns with only 1 unique value
for c in data.columns:
print(c, " num unique values: ", len(data[c].unique()), "\n")
if len(data[c].unique()) == 1:
data.drop([c], axis=1, inplace=True)
submission_data.drop([c], axis=1, inplace=True)
print(data.shape)
print(submission_data.shape)
EmployeeID num unique values: 1340 Age num unique values: 43 BusinessTravel num unique values: 3 DailyRate num unique values: 793 Department num unique values: 3 DistanceFromHome num unique values: 29 Education num unique values: 5 EducationField num unique values: 6 EmployeeCount num unique values: 1 EnvironmentSatisfaction num unique values: 4 Gender num unique values: 2 HourlyRate num unique values: 71 JobInvolvement num unique values: 4 JobLevel num unique values: 5 JobRole num unique values: 5 JobSatisfaction num unique values: 4 MaritalStatus num unique values: 3 MonthlyIncome num unique values: 1134 MonthlyRate num unique values: 1191 NumCompaniesWorked num unique values: 10 Over18 num unique values: 1 OverTime num unique values: 2 PercentSalaryHike num unique values: 15 PerformanceRating num unique values: 2 RelationshipSatisfaction num unique values: 4 StandardHours num unique values: 1 Shift num unique values: 4 TotalWorkingYears num unique values: 40 TrainingTimesLastYear num unique values: 7 WorkLifeBalance num unique values: 4 YearsAtCompany num unique values: 37 YearsInCurrentRole num unique values: 19 YearsSinceLastPromotion num unique values: 16 YearsWithCurrManager num unique values: 18 (1340, 31) (336, 31)
See that there are 9 categorical columns which need to be converted to numerical.
See that there are many numerical columns which need to be binned by quantile.
Dropped columns with only 1 unique value -> can't get any information from those.
Can see that there are no null values, so we do not need to clean out rows or columns containing nulls.
# Binning
int_cols = []
for c in data.columns:
if data[c].dtype == "int64":
print(c, len(data[c].unique()))
print(data[c].describe(), "\n\n\n")
EmployeeID 1340 count 1.340000e+03 mean 1.460265e+06 std 2.494821e+05 min 1.025177e+06 25% 1.237599e+06 50% 1.469862e+06 75% 1.670131e+06 max 1.886378e+06 Name: EmployeeID, dtype: float64 Age 43 count 1340.000000 mean 36.580597 std 9.013072 min 18.000000 25% 30.000000 50% 35.000000 75% 42.000000 max 60.000000 Name: Age, dtype: float64 DailyRate 793 count 1340.000000 mean 799.197761 std 399.333256 min 102.000000 25% 465.000000 50% 796.000000 75% 1153.000000 max 1499.000000 Name: DailyRate, dtype: float64 DistanceFromHome 29 count 1340.000000 mean 9.193284 std 8.141621 min 1.000000 25% 2.000000 50% 7.000000 75% 14.000000 max 29.000000 Name: DistanceFromHome, dtype: float64 Education 5 count 1340.000000 mean 2.924627 std 1.036088 min 1.000000 25% 2.000000 50% 3.000000 75% 4.000000 max 5.000000 Name: Education, dtype: float64 EnvironmentSatisfaction 4 count 1340.000000 mean 2.709701 std 1.099961 min 1.000000 25% 2.000000 50% 3.000000 75% 4.000000 max 4.000000 Name: EnvironmentSatisfaction, dtype: float64 HourlyRate 71 count 1340.000000 mean 65.559701 std 20.335025 min 30.000000 25% 48.000000 50% 65.000000 75% 83.000000 max 100.000000 Name: HourlyRate, dtype: float64 JobInvolvement 4 count 1340.000000 mean 2.717910 std 0.717523 min 1.000000 25% 2.000000 50% 3.000000 75% 3.000000 max 4.000000 Name: JobInvolvement, dtype: float64 JobLevel 5 count 1340.000000 mean 2.051493 std 1.104491 min 1.000000 25% 1.000000 50% 2.000000 75% 3.000000 max 5.000000 Name: JobLevel, dtype: float64 JobSatisfaction 4 count 1340.000000 mean 2.746269 std 1.111328 min 1.000000 25% 2.000000 50% 3.000000 75% 4.000000 max 4.000000 Name: JobSatisfaction, dtype: float64 MonthlyIncome 1134 count 1340.000000 mean 6433.381343 std 4687.058380 min 1051.000000 25% 2870.000000 50% 4876.500000 75% 8038.750000 max 19973.000000 Name: MonthlyIncome, dtype: float64 MonthlyRate 1191 count 1340.000000 mean 14290.377612 std 7166.995911 min 2094.000000 25% 7967.250000 50% 14288.500000 75% 20472.500000 max 26997.000000 Name: MonthlyRate, dtype: float64 NumCompaniesWorked 10 count 1340.000000 mean 2.600000 std 2.472794 min 0.000000 25% 1.000000 50% 1.000000 75% 4.000000 max 9.000000 Name: NumCompaniesWorked, dtype: float64 PercentSalaryHike 15 count 1340.000000 mean 15.168657 std 3.661956 min 11.000000 25% 12.000000 50% 14.000000 75% 18.000000 max 25.000000 Name: PercentSalaryHike, dtype: float64 PerformanceRating 2 count 1340.000000 mean 3.152239 std 0.359386 min 3.000000 25% 3.000000 50% 3.000000 75% 3.000000 max 4.000000 Name: PerformanceRating, dtype: float64 RelationshipSatisfaction 4 count 1340.000000 mean 2.700000 std 1.079858 min 1.000000 25% 2.000000 50% 3.000000 75% 4.000000 max 4.000000 Name: RelationshipSatisfaction, dtype: float64 Shift 4 count 1340.000000 mean 0.808209 std 0.856251 min 0.000000 25% 0.000000 50% 1.000000 75% 1.000000 max 3.000000 Name: Shift, dtype: float64 TotalWorkingYears 40 count 1340.000000 mean 11.222388 std 7.696043 min 0.000000 25% 6.000000 50% 10.000000 75% 15.000000 max 40.000000 Name: TotalWorkingYears, dtype: float64 TrainingTimesLastYear 7 count 1340.000000 mean 2.785821 std 1.263473 min 0.000000 25% 2.000000 50% 3.000000 75% 3.000000 max 6.000000 Name: TrainingTimesLastYear, dtype: float64 WorkLifeBalance 4 count 1340.000000 mean 2.771642 std 0.700007 min 1.000000 25% 2.000000 50% 3.000000 75% 3.000000 max 4.000000 Name: WorkLifeBalance, dtype: float64 YearsAtCompany 37 count 1340.000000 mean 7.070149 std 6.039663 min 0.000000 25% 3.000000 50% 5.000000 75% 10.000000 max 40.000000 Name: YearsAtCompany, dtype: float64 YearsInCurrentRole 19 count 1340.000000 mean 4.272388 std 3.677798 min 0.000000 25% 2.000000 50% 3.000000 75% 7.000000 max 18.000000 Name: YearsInCurrentRole, dtype: float64 YearsSinceLastPromotion 16 count 1340.000000 mean 2.175373 std 3.222376 min 0.000000 25% 0.000000 50% 1.000000 75% 3.000000 max 15.000000 Name: YearsSinceLastPromotion, dtype: float64 YearsWithCurrManager 18 count 1340.000000 mean 4.167164 std 3.581605 min 0.000000 25% 2.000000 50% 3.000000 75% 7.000000 max 17.000000 Name: YearsWithCurrManager, dtype: float64
# Binning
# found manually
cols_to_bin = ["Age", "DailyRate", "DistanceFromHome",
"HourlyRate", "MonthlyIncome", "MonthlyRate",
"PercentSalaryHike", "TotalWorkingYears",
"YearsAtCompany","YearsInCurrentRole",
"YearsWithCurrManager", "NumCompaniesWorked",
"YearsSinceLastPromotion",]
print(data.shape)
print(submission_data.shape)
# uneven 4 groups
for c in cols_to_bin:
try:
data[c+"_Even"] = pd.cut(data[c], 4, labels=False)
submission_data[c+"_Even"] = pd.cut(submission_data[c], 4, labels=False)
except:
print("failed")
pass
print(data.shape)
print(submission_data.shape)
(1340, 31) (336, 31) (1340, 44) (336, 44)
# get dummies from int categorical data
# found manually
already_categorical = ["Education", "EnvironmentSatisfaction",
"JobInvolvement", "JobLevel", "JobSatisfaction",
"PerformanceRating", "RelationshipSatisfaction",
"Shift", "TrainingTimesLastYear",
"WorkLifeBalance"]
for c in already_categorical:
temp_dummy = pd.get_dummies(data[c], prefix=c)
data = pd.concat([data, temp_dummy], axis=1)
sub_temp_dummy = pd.get_dummies(submission_data[c], prefix=c)
submission_data = pd.concat([submission_data, sub_temp_dummy], axis=1)
print(data.shape)
print(submission_data.shape)
(1340, 87) (336, 87)
# get dummies from obj categorical data
for c in data.columns:
if data[c].dtype == "object":
if len(data[c].unique()) == 2:
data[c] = pd.factorize(data[c])[0]
submission_data[c] = pd.factorize(submission_data[c])[0]
else:
temp_dummy = pd.get_dummies(data[c], prefix=c)
data = pd.concat([data, temp_dummy], axis=1)
data[c] = pd.factorize(data[c])[0]
sub_temp_dummy = pd.get_dummies(submission_data[c], prefix=c)
submission_data = pd.concat([submission_data, sub_temp_dummy], axis=1)
submission_data[c] = pd.factorize(submission_data[c])[0]
print(data.shape)
print(submission_data.shape)
# turn label column to binary
labels = pd.DataFrame(pd.factorize(labels)[0], columns=["Attrition"])
(1340, 107) (336, 107)
# Histograms for binned data and dummmy data created
promising_data2 = data[data.columns[31:]]
promising_data2 = pd.concat([promising_data2, labels], axis=1)
for c in promising_data2.columns:
sns.FacetGrid(promising_data2,
hue="Attrition",
height= 5).map(sns.histplot,c).add_legend()
# Feature selection using coefficient weights
estimator = LogisticRegression(max_iter=120)
selector = RFECV(estimator, min_features_to_select=20,
step=1, cv=5)
selector = selector.fit(data, np.ravel(labels))
names = selector.get_feature_names_out()
# See which features are correlated with Attrition
print(data[names].columns)
print(data[names].shape)
Index(['BusinessTravel', 'Department', 'DistanceFromHome', 'Education',
'EnvironmentSatisfaction', 'Gender', 'HourlyRate', 'JobInvolvement',
'JobLevel', 'JobRole', 'JobSatisfaction', 'MaritalStatus',
'NumCompaniesWorked', 'OverTime', 'PercentSalaryHike',
'PerformanceRating', 'RelationshipSatisfaction', 'Shift',
'TotalWorkingYears', 'TrainingTimesLastYear', 'WorkLifeBalance',
'YearsAtCompany', 'YearsInCurrentRole', 'YearsSinceLastPromotion',
'YearsWithCurrManager', 'Age_Even', 'DailyRate_Even',
'DistanceFromHome_Even', 'HourlyRate_Even', 'MonthlyIncome_Even',
'MonthlyRate_Even', 'TotalWorkingYears_Even', 'YearsInCurrentRole_Even',
'YearsWithCurrManager_Even', 'NumCompaniesWorked_Even',
'EnvironmentSatisfaction_1', 'EnvironmentSatisfaction_3',
'EnvironmentSatisfaction_4', 'JobInvolvement_1', 'JobInvolvement_2',
'JobInvolvement_3', 'JobInvolvement_4', 'JobLevel_1', 'JobLevel_2',
'JobLevel_3', 'JobSatisfaction_1', 'JobSatisfaction_4',
'RelationshipSatisfaction_4', 'Shift_0', 'Shift_1', 'Shift_2',
'TrainingTimesLastYear_2', 'TrainingTimesLastYear_3',
'WorkLifeBalance_1', 'WorkLifeBalance_3',
'BusinessTravel_Travel_Frequently', 'BusinessTravel_Travel_Rarely',
'Department_Cardiology', 'Department_Maternity',
'EducationField_Marketing', 'EducationField_Medical',
'JobRole_Therapist', 'MaritalStatus_Divorced', 'MaritalStatus_Married',
'MaritalStatus_Single'],
dtype='object')
(1340, 65)
# Eliminate redundant features
good_data = data[names]
selected_features = mrmr.mrmr_classif(good_data,
np.ravel(labels),
K=20)
print(selected_features)
uncorr_data = good_data[selected_features]
uncorr_sub_data = submission_data[selected_features]
100%|██████████| 20/20 [00:00<00:00, 86.50it/s]
['OverTime', 'JobLevel_1', 'BusinessTravel_Travel_Rarely', 'JobInvolvement', 'Shift_0', 'WorkLifeBalance_1', 'Age_Even', 'DistanceFromHome_Even', 'EnvironmentSatisfaction', 'YearsInCurrentRole', 'MaritalStatus_Single', 'TotalWorkingYears', 'JobSatisfaction', 'JobLevel_2', 'JobInvolvement_1', 'YearsWithCurrManager', 'JobLevel', 'EnvironmentSatisfaction_1', 'TrainingTimesLastYear_2', 'MaritalStatus']
# Save the selected data
fin_data = pd.concat([uncorr_data, labels], axis=1)
fin_data.to_csv("uncorr20_data.csv")
uncorr_sub_data.to_csv("uncorr20_sub_data.csv")
# Feature engineering by making polynomial features
poly = PolynomialFeatures(degree=2)
poly_data = poly.fit_transform(uncorr_data)
poly_sub = poly.fit_transform(uncorr_sub_data)
poly_names = poly.get_feature_names_out()
# print(poly_names)
# print(poly_data.shape)
poly_data = pd.DataFrame(poly_data, columns=poly_names)
poly_sub = pd.DataFrame(poly_sub, columns=poly_names)
# print(poly_data.head())
# Feature selection using coefficient weights
estimator = LogisticRegression(max_iter=120)
selector = RFECV(estimator, min_features_to_select=20,
step=1, cv=5)
selector = selector.fit(poly_data, np.ravel(labels))
names = selector.get_feature_names_out()
print(poly_data[names].columns)
print(poly_data[names].shape)
Index(['OverTime', 'YearsInCurrentRole', 'MaritalStatus_Single',
'YearsWithCurrManager', 'OverTime^2', 'OverTime JobLevel_1',
'OverTime YearsInCurrentRole', 'OverTime JobSatisfaction',
'OverTime EnvironmentSatisfaction_1',
'OverTime TrainingTimesLastYear_2', 'OverTime MaritalStatus',
'JobLevel_1 WorkLifeBalance_1', 'JobLevel_1 Age_Even',
'JobLevel_1 DistanceFromHome_Even', 'JobLevel_1 MaritalStatus_Single',
'JobLevel_1 EnvironmentSatisfaction_1',
'BusinessTravel_Travel_Rarely WorkLifeBalance_1',
'BusinessTravel_Travel_Rarely Age_Even',
'BusinessTravel_Travel_Rarely DistanceFromHome_Even',
'BusinessTravel_Travel_Rarely JobInvolvement_1',
'BusinessTravel_Travel_Rarely JobLevel',
'BusinessTravel_Travel_Rarely TrainingTimesLastYear_2',
'BusinessTravel_Travel_Rarely MaritalStatus',
'JobInvolvement MaritalStatus_Single', 'JobInvolvement JobSatisfaction',
'JobInvolvement JobLevel_2', 'JobInvolvement EnvironmentSatisfaction_1',
'JobInvolvement MaritalStatus', 'Shift_0 WorkLifeBalance_1',
'Shift_0 DistanceFromHome_Even', 'Shift_0 EnvironmentSatisfaction',
'Shift_0 YearsInCurrentRole', 'Shift_0 MaritalStatus_Single',
'Shift_0 JobInvolvement_1', 'Shift_0 YearsWithCurrManager',
'Shift_0 EnvironmentSatisfaction_1', 'Shift_0 TrainingTimesLastYear_2',
'Shift_0 MaritalStatus', 'WorkLifeBalance_1 DistanceFromHome_Even',
'WorkLifeBalance_1 YearsInCurrentRole',
'WorkLifeBalance_1 TotalWorkingYears',
'WorkLifeBalance_1 JobSatisfaction', 'WorkLifeBalance_1 JobLevel_2',
'WorkLifeBalance_1 YearsWithCurrManager', 'Age_Even^2',
'Age_Even EnvironmentSatisfaction', 'Age_Even YearsInCurrentRole',
'Age_Even MaritalStatus_Single', 'Age_Even JobSatisfaction',
'Age_Even JobLevel_2', 'Age_Even MaritalStatus',
'DistanceFromHome_Even EnvironmentSatisfaction',
'DistanceFromHome_Even JobLevel_2',
'DistanceFromHome_Even JobInvolvement_1',
'DistanceFromHome_Even TrainingTimesLastYear_2',
'DistanceFromHome_Even MaritalStatus',
'EnvironmentSatisfaction YearsInCurrentRole',
'EnvironmentSatisfaction MaritalStatus_Single',
'EnvironmentSatisfaction JobInvolvement_1',
'EnvironmentSatisfaction YearsWithCurrManager',
'EnvironmentSatisfaction TrainingTimesLastYear_2',
'EnvironmentSatisfaction MaritalStatus',
'YearsInCurrentRole MaritalStatus_Single',
'YearsInCurrentRole MaritalStatus', 'MaritalStatus_Single^2',
'MaritalStatus_Single TotalWorkingYears',
'MaritalStatus_Single JobLevel_2',
'MaritalStatus_Single JobInvolvement_1',
'MaritalStatus_Single JobLevel',
'MaritalStatus_Single EnvironmentSatisfaction_1',
'MaritalStatus_Single MaritalStatus', 'JobSatisfaction JobLevel_2',
'JobSatisfaction TrainingTimesLastYear_2',
'JobSatisfaction MaritalStatus', 'JobLevel_2 JobInvolvement_1',
'JobLevel_2 MaritalStatus', 'JobInvolvement_1 YearsWithCurrManager',
'JobInvolvement_1 JobLevel', 'JobInvolvement_1 TrainingTimesLastYear_2',
'JobInvolvement_1 MaritalStatus', 'YearsWithCurrManager JobLevel',
'YearsWithCurrManager EnvironmentSatisfaction_1',
'YearsWithCurrManager MaritalStatus',
'EnvironmentSatisfaction_1 MaritalStatus', 'TrainingTimesLastYear_2^2',
'MaritalStatus^2'],
dtype='object')
(1340, 86)
# Eliminate redundant features
good_poly_data = poly_data[names]
selected_features = mrmr.mrmr_classif(good_poly_data,
np.ravel(labels),
K=20)
print(selected_features)
uncorr_poly_data = good_poly_data[selected_features]
uncorr_poly_sub_data = poly_sub[selected_features]
100%|██████████| 20/20 [00:00<00:00, 50.45it/s]
['OverTime JobLevel_1', 'Shift_0 EnvironmentSatisfaction_1', 'DistanceFromHome_Even TrainingTimesLastYear_2', 'JobInvolvement JobSatisfaction', 'OverTime MaritalStatus', 'BusinessTravel_Travel_Rarely Age_Even', 'JobLevel_1 MaritalStatus_Single', 'OverTime', 'Shift_0 JobInvolvement_1', 'YearsWithCurrManager', 'OverTime EnvironmentSatisfaction_1', 'JobLevel_1 WorkLifeBalance_1', 'OverTime^2', 'Age_Even EnvironmentSatisfaction', 'Shift_0 DistanceFromHome_Even', 'JobInvolvement JobLevel_2', 'OverTime TrainingTimesLastYear_2', 'YearsInCurrentRole', 'MaritalStatus_Single EnvironmentSatisfaction_1', 'OverTime JobSatisfaction']
# Generate histograms to check usefulness of features selected
fin_poly_data = pd.concat([uncorr_poly_data, labels], axis=1)
for c in fin_poly_data.columns:
sns.FacetGrid(fin_poly_data,
hue="Attrition",
height= 5).map(sns.histplot,c).add_legend()
# save the final poly data
fin_poly_data = pd.concat([uncorr_poly_data, labels], axis=1)
fin_poly_data.to_csv("uncorr20_poly_data.csv")
uncorr_poly_sub_data.to_csv("uncorr20_poly_sub_data.csv")